MCP Server
AI agents such as ChatGPT and Claude can find and compare Polymarket and Kalshi markets and analyze social sentiment through the MCP server.
The MCP server connects an AI agent, such as ChatGPT, Claude, Grok, Gemini, or a coding agent like Claude Code or Codex, to the research tools on Glypse. MCP, the Model Context Protocol, is a standard way for an AI app to use tools from another service. Once the server is connected, you can ask your agent about prediction markets in your own words. It can find markets on Polymarket and Kalshi, look into an event and how its odds have moved, compare two markets for arbitrage, and read the mood around any subject, then answer in the conversation.
The tools only read data. Your agent can't place an order, move funds, or change your account through them, and Glypse does not execute trades or hold funds.
What you need
You need a free Glypse account and an AI app that lets you add a remote MCP server. Apps give the setting different names, such as a custom connector, a custom app, or a custom MCP server. The feature page has the server URL and the setup steps for each app it covers, with a command or configuration to copy where an app uses one, and notes any plan or workspace permission an app requires.
Create the account before you connect
If you start connecting without an account, the sign-in page offers Sign up, but the connection doesn't carry through sign-up, which ends with confirming your email address. Once your account is ready, start the connection again from your app.
Connecting an app
Connecting an app takes a few minutes. The sign-in happens in your browser, on Glypse's own pages rather than inside the app:
- Create your free account, unless you already have one.
- Add the server in your app, using the server URL and the steps for that app on the feature page.
- Sign in to Glypse when your app opens a browser window. Depending on the app, the window opens when you add the server or the first time your agent uses it.
- Review the request and choose Allow.
- Ask your agent about any market.
The page that asks for your approval names the app and the account you're signed in as, and lists what the app will be able to do, such as looking up markets and social sentiment for you. When the app names the site it comes from, such as claude.ai, the page shows that too. Allow completes the connection, and the app then lists the server's tools. Cancel stops the request without giving the app any access.
What your agent can do
The tools work like the site's four research features, so your agent can answer the questions those pages answer, and each page's manual explains the figures in more detail. You ask in your own words, and your agent picks the tools and fills in what they need from the conversation, such as an event you mentioned or the period you care about.
Finding markets
Your agent can find events on both venues from a name, a topic, or part of a question, as Cross-venue Search does, and can search the resolution rules when a source or threshold matters. It can also screen the whole catalog with the filters of the Market Screener: whole events by category, status, closing time, listing date, volume, and open interest, or individual contracts by their current chance and their move over the past day. A question can start from a hunch, such as “I think the Bears make a real run this year. What can I bet on?”, and your agent finds the markets that would pay if it happens.
Examining an event
Given an event, whether from an earlier answer, a research page link you paste, or a name it looks up first, your agent can read what the event details show: each market's current chance and change over the past day, the volumes and open interest, key dates, and the venue's resolution sources. It fetches the full rules when you ask how a market settles, and it can trace how the odds have moved, for the event's main markets or one market you name, over the ranges of the site's price chart. “What exactly does this market settle on?” brings back the rules, and “How has it moved this month?” the price history.
Comparing two markets
Starting from one event, your agent can list the most related events on both venues: the same question listed on the other venue, and neighbouring questions on either. These are the suggestions under Relevance in the Arbitrage Scanner and on a research page's Related events card. It then compares two markets the way the scanner does, with the cost of one contract in each for every pairing of Yes and No, the payout and gain or loss under each settlement outcome, and how the gap between their prices has moved. Asked about a Polymarket event, a question such as “Is this cheaper on Kalshi?” covers both steps.
When an event has several markets and the question names none, the comparison uses the event's leading market and says so. Naming the candidate or price threshold you mean compares that market instead.
Analyzing social sentiment
Your agent can analyze the posts behind the Social Intelligence dashboard for any subject, with the dashboard's measurements: post volume over time, the sentiment score and its balance of positive, neutral, and negative posts, the mix of emotions and topics, and the most-mentioned hashtags, cashtags, accounts, and sites. It can also compare several subjects over one period, count the posts about a subject before running a full analysis, and read the newest posts with their labels, or only the negative or angry ones. “How's the mood on Bitcoin this week?” draws on the analysis, and a follow-up such as “What flipped it?” on the posts behind it.
An analysis covers a preset period ending now or a pair of dates, and can be limited to topics such as Technology or Business. Mentioning your time zone lets the daily figures follow your own calendar days.
The tools in your app
Apps list each tool by its name, as shown below, or by a plain title such as Search events. You don't need to name a tool yourself, since your agent chooses them, but the table shows what each one does and the part of the site it mirrors.
| Tool | What it does | On the site |
|---|---|---|
search_events | Finds events on both venues from a name, a topic, or part of a question, optionally searching the resolution rules too. | Cross-venue Search |
screen_events | Lists whole events filtered by category, status, timing, volume, open interest, event structure, and keywords. | Market Screener, Events view |
screen_markets | Lists individual contracts filtered by current chance, 24-hour move, volume, timing, category, and keywords. | Market Screener, Markets view |
list_market_categories | Gives the categories and subcategories that organize both venues' markets together. | Market Screener categories |
suggest_screener_keywords | Checks how the catalog spells a name or term, since screener keywords match whole words exactly. | Market Screener keywords |
get_event | Returns one event with its markets, current chances, trading figures, dates, and resolution sources, with the full rules on request. | Event details |
get_event_price_history | Shows how the odds of an event's main markets have moved over a chart range. | Event details, price history |
get_market_price_history | Shows how one market's odds have moved over a chart range. | Event details, price history |
find_related_events | Lists the events most related to one event: the same question on the other venue and neighbouring questions. | Arbitrage Scanner, Relevance |
compare_markets | Compares two markets: the cost of each pairing of sides, the payout and gain or loss under each outcome, and the two price histories. | Arbitrage Scanner |
analyze_social_sentiment | Analyzes one subject's post volume, sentiment, emotions, topics, and most-mentioned tags, accounts, and sites over a time range. | Social Intelligence dashboard |
compare_social_sentiment | Puts several subjects side by side over the same time range. | Social Intelligence dashboard, Compare view |
read_social_posts | Returns the newest matching posts with their labels, optionally only those with a chosen sentiment or emotion. | Social Intelligence dashboard, Recent posts |
count_social_posts | Counts the posts a query matches without the full analysis. | Social Intelligence dashboard, a query's hits count |
check_social_query | Says whether a query is accepted and describes in plain words which posts it matches. | Query Syntax |
get_usage | Reports your remaining calls and units, and when each resets. | Your account |
Reading the answers
Your agent explains what it finds in its own words, using the same terms as the site's pages.
Events and price history
Chances are Yes prices read as probabilities, as on the site, and some markets use the venue's own side names, such as Up and Down, in place of Yes and No. Your agent can link each event to its research page, where its briefing and social section can be read. Prices come from the same regularly refreshed data as the site's pages, so they can trail the venues slightly.
For a price history, your agent can say where the odds started and ended over the range, how many percentage points they moved, and their highest and lowest values. For a recently listed event, it covers the history that exists.
Arbitrage comparisons
A comparison gives both markets' chances, the gap between them in percentage points, and their sum. It then reads the pair in two ways, each priced at whichever of its two pairings costs less:
| Reading | Position | Fits a pair where | Payout per pair of contracts |
|---|---|---|---|
| Hedge | Yes on one market and No on the other | Both markets should settle the same way, as with one question listed on both venues | $1 when they settle alike; $2 or nothing when they split |
| Same side | Yes on both, or No on both | Exactly one of the two outcomes can happen, as with the two teams in a final | $1 when exactly one settles that way |
Your agent leads with the reading that fits the pair: the same-side reading for two markets of one winner-take-all event, and the hedge for any other pair, with the other reading as an alternative. It can also say how far the gap has ranged over the chart range and where it stands now. The Arbitrage Scanner manual explains the cost and payout grids these readings come from.
The rules decide whether a pair is an arbitrage
A reading is a gain only under the settlement outcomes the two markets' rules allow. Markets with similar titles can still differ in deadline, threshold, or resolution source, so asking your agent to read both sets of rules settles which reading applies. The figures are estimates from displayed prices: they leave out bid/ask spreads, venue fees, how many contracts are available at that price, and voided or partial settlements.
Social analyses
An analysis states how many posts matched, the average per day or week, the busiest period, and whether volume is increasing, decreasing, or stable. The sentiment score runs from -100 to +100 and measures the balance between positive and negative posts. It comes with one of five levels, from Very Negative through Neutral to Very Positive, and with a trend that is improving, declining, or stable. The most positive and most negative day or week are named as well. Emotions and topics are reported as shares of their labels, led by the most common. A comparison gives each subject's post count, score, and level, then names the most and least positive subjects and the leading emotion and topic.
The Sentiment reference gives each level's score range and explains how the trend is calculated, and the Emotions and Topics references describe their labels. A label describes the tone of a post as a whole: a disappointed supporter's post reads as negative, for example. Reading the posts behind a figure shows what people are reacting to before the figure is taken as a view on how a market will settle.
Social queries
Your agent writes the searches for social posts itself, in the query language of the dashboard's Write queries mode. A query names a subject the way people write about it: every word in it must appear in a post, and a pipe separates alternatives, as in Nvidia | NVDA for a company and its ticker. The Query Syntax reference explains the rest of the operators.
You can ask your agent to show its queries and check them before relying on the figures, and a query copied from the dashboard runs the same way. Your agent can also have a query read back in plain words before running it, to confirm it covers what you meant. Queries your agent runs are not added to your query history or bookmarks on the dashboard.
Usage limits
Each account has two limits on the server. One caps how many calls your agent can make in a minute, and every call counts the same, whatever the tool. The other is a daily allowance measured in units: each tool uses its own amount, a social comparison counts each subject it includes, and tools that look up no data, such as checking a query or reporting your usage, are free. The daily allowance starts with the first call that uses it and resets a day after that call, not at midnight.
You can ask your agent how much of the allowance is left and when it resets.
When a limit is reached, your agent gets a message instead of data. After too many calls in a short time, the message says how many seconds to wait, and when the daily allowance is used up, it says how long until the allowance resets. A call turned away by the per-minute limit uses no units.
Managing connected apps
Account Settings, in the account menu, lists every app you have allowed under Connected apps in its Security card, with the site it comes from when it names one and the date it was connected. The trash-can button beside an app signs it out; pointing at the button shows Sign out.
Signing an app out stops it from renewing its access. Access it already holds keeps working until it expires, after which the app needs a new sign-in. Sign out of all sessions, at the bottom of the same card, signs out every device and every connected app, including the browser you are using. Removing the server in the app's own settings stops the app from using it, while signing it out under Connected apps ends its access on the Glypse side.
Some apps stay connected on their own, while others ask you to sign in again from time to time. An app that stays connected asks for that on the approval page, where the list includes staying connected without asking you to sign in.
When something goes wrong
The connection does not finish
If the browser page says the request is no longer valid, the request expired or arrived incomplete, and starting the connection again from your app opens a fresh one. If it says the app couldn't be connected, it asks you to sign out of Glypse in that browser, sign back in, and start again from your app. A connection started before you created your account also needs to be started again.
The app asks you to sign in again, or its tools stop working
The app's access has ended, either because the app doesn't stay connected on its own or because it was signed out under Connected apps. Signing in again from the app, or reconnecting the server in its settings, starts a new connection.
Related pages
Server URL and setup steps
The address to add in your app, with setup steps for ChatGPT, Claude, Grok, Gemini, coding agents, and more.
Cross-venue Search
How events are matched by name, topic, and rules wording, and what the event details show.
Market Screener
The filters, categories, and figures behind screening whole events and individual contracts.
Arbitrage Scanner
How the cost, payout, and gain or loss of a pair of contracts are calculated.
Social Intelligence
How the dashboard's queries, measurements, and comparisons work.
Query Syntax
The operators that decide which posts a query matches.